Abstract
Oil and gas wells leakage is a major concern due to the associated risks. Potential issues include habitat fragmentation, soil erosion, groundwater contamination, and greenhouse gas emissions released into the atmosphere. An estimated 2 million abandoned oil and gas wells are believed to be leakage. Proper Plug and Abandonment (P&A) operations are required to ensure these wells are correctly disposed of from their useful operational life. This study aims to build an uncertainty evaluation tool to statistically classify the risk of a well from leaking based on their well information (age, location, depth, completion interval, casings, and cement). Data consists of leakage reports and available well data reports from Alberta Energy Regulator (AER) in Canada. Multiple preprocessing techniques, including balancing the data, encoding, and standardization, were implemented before training. Multiple models that included Naïve Bayes (NB), Support Vector Machine (SVM), Decision Trees (DT), Random Forest (RF), and K-Nearest Neighbors (KNN) were compared to select the best-performing for optimization. RF outperformed the other models and was tuned using hyperparameter optimization and cross-validation. The final model's average accuracy was 77.1% across all folds. Multiple evaluation metrics, including Accuracy, Confusion Matrix, Precision, Recall, and Area Under the ROC Curve (AUC), were used to assess the model and each class against the rest. Feature importance showed an even distribution across the different features used. The model presented in the study aimed to classify wells and label the leakage risk based on the well information associated with its components. This risk evaluation tool could help reduce gas emissions by 28.2% based on the results obtained. This tool can classify the wells to speed the selection process and prioritize wells with higher leakage risk to perform P&A operations and minimize emissions.
